Expanded version of my talk at AIBoomi Annual '26.
Last week Shantanu Narayen stepped down from Adobe. Eighteen years. Best quarter ever. Every bet right. Even the AI bet felt almost right. Business model was the gaping hole.
Did not matter. Markets price the narrative of the future, not the performance of the past.
When I was at Intuit, my CEO would go to Shantanu to learn how to move from desktop to cloud. He was the reference manual for business model transition.
But looks like the board said: your experience is not an asset anymore, it is a liability.
If Shantanu Narayen is in this position, then who are we. At first this was terrifying.
Upon further reflection I found it the most relieving.
If Shantanu, world's best enterprise CEO of the last 20 years, one of the best capital allocators, is in this position then who are we.
This gives us permission, relaxation to say we are at a new starting point.
SaaS Is the New COBOL
SaaS gave us a great business model. Gave India an opening batting role in becoming a product nation.
Before AIBoomi, SaaSBoomi. Before that, SaaSX. I was in that petri dish with the SaaSBoomi co-founders, painting the picture of how big SaaS could be.
Three years ago I stood on a similar stage and said trillion-dollar SaaS economy. Sixty-six billion-dollar-ARR companies at 15x. When one or two break the barrier, dominoes fall.
That story was beautiful. But is wrong now.
SaaS is the new COBOL. $3T in transactions. Still pays well. No Gen Z wants to work there. Forget Gen Z, no millennial wants to work there. SaaS went from the future to the furniture. Antique.
It is like Nandan said at his annual a few weeks ago, the world around changed but our mental models did not change fast enough. They are broken
Broken mental models aren't fixed by better playbooks. They're fixed by better questions.
I want to leave you with 3 questions that I found useful and you can use them on Monday with in your discussions.
Dam Breaking Points
Last year at this conference, a founder was the session manager for my talk. He grilled me like no one else. We revised the talk four times. Five questions on SaaS to AI pivot. What breaks. What survives. What's worth building next.
He had built a SaaS company to multi-million ARR. Bootstrapped.
Few months after that session, he started something else. I had told the room founders they are better off doing a 200% pivot. A hundred-plus founders heard that. This founder shifted
He started a voice AI healthcare startup.
Last Sunday he pinged me. Six months in. What took four years in enterprise sales at his previous SaaS company took six months at the healthcare startup. Healthcare. Enterprise deals. Multiple people in the room to say yes. Not AI PLG where a developer swipes a credit card and the thing grows. The longest sales cycle in software.
Fastest velocity I had seen. He said the buyers weren't evaluating anymore. They were desperate. Billions of patient calls a year. Millions in staff shortage. Huge turnover. Problem got so bad that when the technology got good enough, a dam broke.
Other broader examples of dam breaks in last twelve months. Claude Code. Voice Agents. OpenClaw.
How do you know when it's happening. People doing things they literally couldn't do before. Non-target ICP users showing up. He didn't decide to go build in healthcare. One thing led to another, his healthcare customers found him. Day 30 bigger than Day 1.
If one unit of progress is not leading to ten units of outcome, you are not working on the right spot.
Build your sidecar. You need someone who can tell you about a dam break six months in advance. The friend shouting about Claude Code in July, not December 2025. Don't pick the PhD still in shock that pre-2020 PhD became irrelevant. Pick the person who sees around corners.
Ka-Ching Products
We are doing a disservice calling Claude Code a dev tools category. People started calling it catnip for programmers. That herb around which cats go crazy. Soon it was clear it is far more addictive.
But Claude Code works because it half works. You put in credits, pull a lever. Sometimes magic, sometimes garbage. Tweak it, load more credits, pull again.
BF Skinner studied this. Variable rewards create addiction. Slot machine doesn't pay out every time. That's why you keep pulling.
You are not building a wow moment. You are baking a digital cocaine.
You're building a casino. First for humans, then for agents. Ka-ching.
But Ka-Ching is necessary, not sufficient.
The turkey story. Nassim Taleb. Black Swan. Day 1 to day 1,000, the turkey master is great. Feeds you. Revenue beautiful. Day 1,001 is Thanksgiving. There is no turkey.
Paul Buchheit applied it to AI startups. Growing fast building workflows around today's models. Future models won't need the workflows.
I worry about the newly minted fastest-to-$100M-ARR stories. They've cracked the Ka-Ching. You can prove $100M. The gobbler still comes.
Attack of the clones. If something works, a hundred replicas get created in days and weeks. Platform gobblers. You show $100M is possible, the platform builds your feature into their next release.
I am in love with Claude Code. Addicted. But it is a venus flytrap. Anthropic will suck your soul and spit you out. You don't own your model. You are not owning your destiny.
ChokeNeck
Gold rush analogy is the most useless mental model in AI. Picks and shovels.
Helicone. Koyeb. Langfuse. All acquihired. Shovels got bought for parts.
Bun got ~$900M. Metronome got ~$800M. They weren't shovels. They were chokepoints. Everything had to flow through them.
People push back. Oh Rajan, retrospective analysis is easy. Just jump in. See growth, stick to it.
Look at the $2 trillion wipeout. Terminal value is the ability to protect cash flow, not generate it. When the ability to protect evaporates, the multiple collapses.
How you grow is a Ka-Ching question. How you protect is a ChokeNeck question.
And the chokepoint moves. Five years ago, talent. Three years ago, training data. Two years ago, the model. Today, inference cost. Harness engineering. Orchestration.
You don't have your inference in your control, you are dead meat.
Look at what you built last quarter that you didn't plan to build. The infrastructure piece you needed and couldn't buy. If three other companies in your space also built it, that's the primitive.
Stop building moats. Find the narrow passage everything flows through. Sit there.
To recap.
Question your mental models.
Recognize the dam breaks. Build a casino. Sit at the chokepoint.
If Microsoft with bad products can refound and survive every shift in the last 50 years then you, a great builder in the audience, can refound it.
Slides from the talk: view on Google Slides.